Re: First failiure of the "fixed" odbc interface on Linux

[Flemming Frandsen]

Mensing, Joerg wrote:
> [EINTR might be the cause of trouble]
> Hi, sounds very much like the reason for the instability.
> Current LINUX pthread library creates lots of 'spurious' signals,
> which are not handled correctly.

Does the client runtime use threads, because I've never seen one show up in top?


> SAPDB client runtime simply expects, that signals are
> used only for killing a running program and not for internal
> dispatching of pthreads. We have to wrap all the system calls, to
> have a stable multithreaded client runtime on LINUX. Not a small job,
> but a necessary one to get it stable.

Well, not handling EINTR by re-trying the system call is always an error IMHO, as you'ld get the same error if the application uses signals to do some inter-process synchronization, something that's still popular.

Some user might suspend a process with ^Z, for example and that is true for any unix, not just Linux with fscked pthreads:-)

Some platforms implement sleep() using SIGALRM, so that could lead to trouble as well...


>> There is also no trace of trying to report the actual
>> errorcode (errno), something that would make it a lot easier to
>> debug this problem...
>
>That is true. I currently can only suggest to use 'strace' to pin down >the problem...

Hmm, strace does have some sideeffects timing-wise, so a bug may not show up while stracing...

--
Regards Flemming Frandsen - http://dion.swamp.dk
PartyTicket.Net co founder & Yet Another Perl Hacker

_______________________________________________
sapdb.general mailing list
[EMAIL PROTECTED]
http://listserv.sap.com/mailman/listinfo/sapdb.general